Oxford university press how the br ain evolved language

Figure 8.5. 
Anatomy for serial learning with self-inhibition. 
enough for the seriating STM-LTM gradient (figure 8.4) to develop. Later, 
during performance, the rehearsal wave could be turned off and “suicide” 
inhibition enabled to prevent perseveration. The problem with this solution is 
in finding the suicide loops. If they are endogenous to the cerebrum, they vio­
late the principle of on-center off-surround cerebral organization. There is, 
however, another way to draw the suicide loops of figure 8.5, and it involves 
the cerebellum. 
In the beginning of a serial performance, assume frontal neocortex forms 
a broad motor plan. In sentence 8.1, for example, it may be assumed to plan a 
word like runs. 
The dog runs down the street. 
(8.1) 
We assume this because in Broca’s aphasia, inflections like the -s of runs often 
fail to form. Similarly, in dysarthrias, motor disabilities affecting speech, which 
may result from lesions near Broca’s area, the individual phonemes of a word 
like run may also be misplanned. By this analysis, the cerebrum passes a se­
quence of commands such as ([r]-[
]-[n])-[z] to the terminal articulatory 
musculature. En route, however, these signals are modulated by the cerebel­
lum, whose task it is to coordinate such fine motor activity. 
Several things happen in the schematic cerebral-cerebellar circuit of fig­
ure 8.6. Descending cerebral motor commands are passed down the pyrami­
dal tract (PT), but axon collaterals carry copies of these commands to brain 
stem relay points like the pontine nucleus (PN) and the inferior olivary complex 

ONE
, 
TWO
, 
THREE 
•  129 
Figure 8.6. 
Cerebellar modulation and feedback. (Allen and Tsukahara 1974. 
Reprinted by permission of the American Physiological Association.) 
(IO), where the descending motor commands are joined by ascending, prop-
rioceptive signals. A loop can be traced through these nuclei, across the mid­
line to the dentate nucleus (DE) of the cerebellum, and back to motor cortex 
through the ventrolateral thalamus (VL). The signals in this loop are excitatory 
and as such generate on-center feedback. By itself, this loop could cause the 
cerebral motor command to perseverate: in our example of runs, it might pro­
duce r-r-r-r. . . . 
Mossy fibers from the pontine nucleus and climbing fibers from the infe­
rior olive cross the midline and project directly to the cerebellar hemispheres. 
(Unlike the cerebrum, the right half of the cerebellum controls the right half 
of the body.) When articulation of [r] is sufficiently accomplished for articula­
tion of [
] to begin, the cerebellum signals this fact to the dentate nucleus. 
Now it happens that the main, large, output neurons of the cerebellum, the 
Purkinje cells (P.C.), are inhibitory. This means cerebellar output serves to break 
the perseveratory on-center feedback loop. Purkinje cells are our “suicide 
cells.”
4
 (The reader should experience a certain sense of déjà vu in this account. 
The circuitry of figure 8.6 is essentially the same circuitry as the six-celled brain 
we evolved in figure 2.9, and it solves essentially the same problem: how to stop 
a motor command. A car needs brakes as much as it needs an engine.) 
Now let us return from the issue of perseveration in serial performance to 
look again at the serial-learning process described in figures 8.5. If cerebellar 

130  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
inhibition is active when one is presented during learning, then x
1
 will not in­
hibit x
2
, z
1
 will not become greater than z
2
, and serial order will not be learned! 
Clearly, cerebellar inhibition must be shut off during learning, but turned on 
during performance. This can be done in several ways. 
First, cerebellar inhibition can be overridden by competition from cere­
bral excitation, that is, by attention. For example, when learning to play a 
B-major arpeggio, one might visually augment the nascent B-D#-F#-B motor 
plan by looking at a score. Simultaneously, one might repeat these note names 
aloud. Such additional inputs could override cerebellar inhibition during learn­
ing. Later, these supplemental excitatory inputs can be turned off for an “au­
tomatic” performance. A second available mechanism, which does not depend 
upon such cognitive crutches, is tempo. 
Tempo 
The tempo with which any serial behavior is performed, be it a word or an 
arpeggio at the piano, can be broadly controlled by nonspecific arousal (NSA). 
As we generally increase nonspecific inputs to an ordered gradient like the x
i
’s 
in figure 8.5, each site reaches threshold and fires sooner, with the result that 
the entire sequence is performed faster. At slower tempi, x
i
 has more time to 
inhibit x
i+1
. Thus, slow tempi maximize the slope of the serial order gradient 
during learning. 
The climbing fibers which arise from the inferior olivary complex of the 
medulla to excite the Purkinje cells are one often-noted source of tempo sig­
nals in the cerebellum. It is unclear to what extent this set of inputs can be 
brought under conscious control or otherwise manipulated during learning, 
but as the tempo of these nonspecific inputs decreases, excitation of the 
Purkinje cells will decrease and cerebellar inhibition of the cerebral motor plan 
will decrease. Thus, slow tempi can also increase the LTM gradient across the 
x
i
 field during learning. 
It seems normal that a combination of the foregoing mechanisms operates 
when one learns to play a B-major arpeggio. At first the motor plan is prac­
ticed very slowly. This allows x
i
 sites corresponding to B-D#-F#-B to achieve a 
steep order gradient. Eventually, the gradient is copied into LTM, and the pia­
nist no longer has to look at the music or say “B-D#-F#.” It is then sufficient to 
simply activate the “arpeggio array” x
j
, and the appropriate fingering is elic­
ited “automatically.” As the tempo increases, the cerebellum activates more 
quickly and keeps the fingering of the arpeggio coordinated by deperseverating 
each note more quickly. 
Stuttering 
A similar analysis can be applied to the learning and fluent performance of 
the serial phonemes of a word. But if during performance the cerebral tempo 

ONE
, 
TWO
, 
THREE 
•  131 
is much faster than the cerebellar tempo, the cerebellum cannot inhibit the 
cerebral motor commands fast enough or strongly enough, so the commands 
perseverate and the speaker stutters. A second kind of stuttering might arise if 
cerebellar inhibition is too strong or the cerebellar tempo too fast. In this case, 
the motor plan can never get started. 
The preceding model is also convergent with data on ataxic dysarthria due 
to cerebellar disease (Schoenle and Groene 1993).
5
 In particular, Kent et al. 
(1979) spectrographically examined five subjects with degenerative cerebel­
lar disease. Some 50% of the phonetic segments they measured exceeded 
normal durations by more than two standard deviations. Figure 8.7 displays such 
lengthening for [p] and [k]. 
It is normal for speech to be slowed by a wide range of neurological disor­
ders which affect language. In the trivial case, the afflicted speaker simply slows 
down in a conscious response to his own difficulty in speaking. But unlike other 
syndromes, cerebellar ataxic dysarthria is especially characterized by the length­
ening of normally brief segments such as unstressed vowels, lax vowels, and 
consonants in clusters. The durations of these features and voice onset time 
are not normally under conscious control, and so indicate, as Kent et al. also 
intimated, that the cerebellum fine-tunes motor speech performance by the 
termination of cerebral motor speech commands.
6 
Dipole Rhythm Generators 
Suicide loops can also provide the circuitry for dipole rhythm generators (Ellias 
and Grossberg 1975). Indeed, our first vertebrate brain in chapter 2 (figure 
2.9) evolved a cerebellum for the very purpose of creating rhythmic movement. 
Vertebrate serial behavior is quintessentially rhythmic serial behavior. We can 
Figure 8.7. 
Lengthening of [p] and [k] by cerebellar ataxics (open circles) and 
normal controls. (Kent et al. 1979. Reprinted by permission of the American 
Speech-Language-Hearing Association.) 

132  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
think of the cerebellum as the “master suicide loop” and the olivocerebellar 
circuit as a kind of “master clock” so long as we recognize that there are actu­
ally many suicide loops, many clocks, and many rhythms in the vertebrate brain 
and in the vertebrate body. Among these are the heart rhythm, the respiratory 
rhythm, the circadian (sleeping/waking) rhythm, the walking rhythm, and the 
rhythms of language, to which we now turn in chapter 9. 

ROMIET AND JULEO
•
133
• 
N 
I 
N 
E
• 
Romiet and Juleo
For never was a tale of more woe, 
Than this of Romiet and her Juleo. 
In chapter 8 we saw how serial order could be stored and retrieved by a paral­
lel brain. We noted, however, that there were limits upon serial learning and 
performance. In particular, we noted that series of more than four items seem 
to exceed our immediate memory span, forcing us to unitize long lists as a list 
of sublists. 
In music, the number of beats per measure rarely exceeds four. When it 
does so, as in a jig or a slip jig, it is usually unitized into two or three subgroups 
of three beats each. The same is true of English. A word like recíprocate has four 
syllables. But as soon as we go to five syllables per word, English words divide 
themselves with a secondary “downbeat.” For example, when recíprocate (four 
syllables) becomes rèciprócity (five syllables), a second downbeat appears. To 
make matters more complicated still, the downbeats “move” to different syl­
lables, and even the sounds of the vowels within the syllables change. In 
music, it is also true that measures of four beats commonly subdivide into two 
groups of two beats. Similarly, most English words of four syllables (and many 
of three syllables) also divide themselves into two beats. Thus óperate (three 
syllables) becomes òperátion (four syllables, two beats). 
In 1968, Chomsky and Halle published The Sound Pattern of English. SPE 
was a remarkable book insofar as it brought considerable order to previously 
confused accounts of English stress patterns. It accounted for stress alterna­
tions like reciprocate-reciprocity by postulating an underlying, abstract, lexical 
representation in which the vowel qualities of words like reciprocity were marked 
as either tense or lax. The stress pattern of a word could then be derived from 
this underlying representation of vowel qualities. 
And in 1966, Brown and MacNeill published a paper entitled “The ‘Tip-of-
the-Tongue’ Phenomenon” (TOT). They presented subjects with the defini­
tions of unfamiliar words. When subjects found the word was “on the tip of 
their tongue” but not quite yet definitely identified/recalled, they were in­
133 

134  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
structed to write down the successive, nearly right words which came to mind. 
Thus, when the definition was “southeast Asian sailing vessel” and the target 
word was sampan, the word boat might have been an expected TOT response. 
Instead, what Brown and MacNeill found was that words like salmon or sump 
pump were more common than words like boat or sailing vessel. That is, sound-
alike words, especially words beginning with the same initial phoneme as the 
target, were retrieved before words of like meaning. But among sound-alike 
words, words with a similar stress pattern to the target were recalled most fre­
quently of all. 
SPE could not account for this TOT result, in which stress pattern seemed 
to be accessed before any underlying vowels. By the mid-1980s, dissatisfaction 
with the SPE account had become widespread on other grounds as well, and 
the description of English stress in terms of metrical phonology (Liberman 1979; 
Selkirk 1984; Pierrehumbert 1987; see Goldsmith 1993 for later developments) 
became widely accepted. 
Our account will support the general thrust of these metrical analyses. 
Words have feet, just as poets have always claimed. These feet may usually be 
associated with certain segmental and featural patterns of a word, but they are 
not serially derived from such patterns. Rather, the stress patterns of words exist 
in parallel with phonetic patterns (in the later jargon of metrical phonology, 
they exist in parallel tiers). This is why stress patterns, initial segments, and 
meaning can all be activated independently in the TOT phenomenon. In this 
chapter, we will see how these “tiers” of metrical phonology are ultimately and 
universally based on cortical cytoarchitecture. 
Spoonerisms 
As we noted in chapter 1, spoonerisms are named after the mal mots of Rev. Wil­
liam Archibald Spooner (1844–1930), Fellow and Warden of New College, 
Oxford. On one occasion, meaning to address a group of workers as “sons of 
toil,” he supposedly instead called them “tons of soil.” In point of fact, 
Dr. Spooner probably uttered only a handful of the many “spoonerisms” at­
tributed to him by his Oxford undergraduates. Hardly an idiosyncratic quirk 
of Dr. Spooner, spoonerisms are ubiquitous and extremely easy to produce. 
What attracts our interest here is not that they are bizarre, but rather that they 
are so very natural. 
Karl Lashley seems to have been the first to recognize that spoonerisms 
were more than just a joke. Lashley (1951) noted that spoonerisms were dev­
astating to serial theories of behavior. To illustrate his argument, consider the 
movements necessary to account for Lashley’s example, “To our queer, old 
dean”: 
↓ 
To our d ear old qu een.
 (9.1) 
↑ 

ROMIET  AND  JULEO 
•  135 
Example 9.1 looks simple on paper, but if one truly believes that the brain 
processes the sounds of 9.1 like so many pieces of movable type, then how does 
it replace the d with qu before it gets to the qu? And where does it put the d 
until it comes to (qu)een, and how does it remember to insert the d when it 
gets there? And if this is easy to explain, then why aren’t there spoonerisms 
like *9.2 or *9.3? 
*dear old queen our 
(9.2) 
*our dean old queer 
(9.3) 
Considerable research into the errors of otherwise normal speech followed 
Lashley’s questions (Chomsky 1972, 3; Fromkin 1973, 1977; Cutler 1982; Garrett 
1993). 
Romiet and Juleo 
One of my more infamous spoonerisms occurred when I was reading the part 
of Chorus in Shakespeare’s Romeo and Juliet. Coming to the closing couplet, I 
dramatically closed the book and recited, 
For never was a tale of more woe, 
(9.4) 
Than this of Romiet and her Juleo. 
Figure 9.1 models the cerebral organization by which I propose to excuse my 
metathetic performance. Let us call this neural system “Spooner’s Circuit.” In 
it, four major linguistic fields are identified corresponding to four levels of unit­
ized motor plans: phrase, word, foot, and syllable. Every phrase unitizes some “magic 
number” of words (n 
≤ 4), and every word unitizes some magic number of feet. 
Every foot consists of two feet (or “beats”): a “left” foot (or “downbeat”) and a 
“right” foot (or “offbeat”). Every beat unitizes a magic number of syllables. For 
simplicity, we will treat each unit as a dipole, and for concreteness, the reader 
may wish to imagine that these plans are located in concentric rings emanating 
rostrally from Broca’s area. At the top level of figure 9.1, the phrases Juliet and 
Romeo and Romeo and Juliet exist in dipole opposition. (The name of the play is 
Romeo and Juliet, but Shakespeare ended the play with the phrase Juliet and her 
Romeo.) At the center of figure 9.1, but somewhat offstage in vivo, a dipole rhythm 
generator oscillates between the left foot and the right foot. 
The boldface entries in tables 9.1–9.4 indicate which pole of each level of 
figure 9.1 is active at times t
1
–t
4
. At t
1
 (table 9.1), the usual phrase, Romeo and 
Juliet, its first word, Romeo, and Romeo’s “left foot,” Rom, were all activated. So 
Rom was output. 
At t
2
 (table 9.2), the rhythm generator shifts the system to the “right foot.” 
Normally, this would activate the eo syllables of Romeo. But at this instant, I re­

136  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
Figure 9.1. 
A Spooner circuit for Romiet and Juleo. 
alized that my plan wasn’t going to rhyme! A burst of nonspecific arousal (NSA) 
rebounded my phrase-level first dipole: I switched to the phrase Juliet and Romeo, 
under which Juliet is the first word preferred by long-term memory (LTM). But 
I couldn’t say Jul: I was “on my right foot.” Instead, out came iet. 
Next, at t
3
, (table 9.3), the system shifted to the left foot. Juliet was still ac­
tive at the word level, so I said Jul. 
Finally, at t
4
, (table 9.4), the foot changed to the right foot, Juliet was 
deperseverated, and the word-level dipole rebounded. Romeo was selected at 
the word level, and eo was selected at the foot level. I said eo. And thus ended 
the tragedy of Romiet and Juleo. 
Lought and Thanguage 
A greater tragedy than Romiet and Juleo has been the confusion of thought and 
language. In the spoonerism lought and thanguage—or in pig latin, for that 
matter—the metathesis cannot be driven by metrical feet in the same manner 
as Romiet and Juleo: thought has only one foot, and only phonemes—not entire 
TABLE
 9.1. 
Output of figure 9.1 at t
1
. 
Phrase 
Juliet Romeo 
Romeo Juliet 
Word 
Juliet 
Romeo 
Foot (L) 
Rom eo 
Output 
“Rom . . .” 

ROMIET  AND  JULEO 
•  137 
TABLE
 9.2. 
Output of figure 9.1 at t
2
, after 
nonspecific arousal. 
Phrase 
Juliet Romeo 
Romeo Juliet 
Word 
Juliet 
Romeo 
Foot (R) 
Jul iet 
Output 
“Romiet . . .” 
syllables—are metathesized. So an additional rhythm generator must be pos­
tulated to drive this spoonerism, as in figure 9.2. 
The operation of this circuit is similar to that of figure 9.1 except that at 
the syllable level, each syllable must be divided into an onset and a rhyme con­
trolled by a distinct, syllabic rhythm generator. We presume that at t
1
 the speaker 
intends to say language. Then for some external reason, the word dipole is re­
bounded at t
2
, just after [l] is produced. As a result, the thought word plan is 
forced active, but the syllabic rhythm generator has switched from onset to 
rhyme: [
ɔt] is output at t
2
. 
With the completion of a syllable, the foot dipole rebounds to the offbeat, 
and a morphological beat (M; and) is output at t
3
. The foot dipole rebounds back 
to the downbeat, but thought and /
θɔt/ still have not been deperseverated. With 
the onset pole of the syllable dipole active, [
θ] is output at t
4
. 
At t
5
, /
θɔt/ and thought are finally deperseverated and the motor plan 
switches back to L. With the rhyme pole of the syllable active, [æ
Ω] is gener­
ated at t
6
 (not diagrammed). At t
7
, the second syllable of language becomes 
active. The onset pole of the syllable activates the learned serial order gradi­
ent [g
w
I
d
á]. In this and the following rhyme, figure 9.2 details the level of cer­
ebellar deperseveration. The sounds [g] and [
w
] are output and deperseverated 
under tight cerebellar control at t
7
. No rhythmic dipole is posited at this level 
of the motor plan or performance, and elements at this level cannot be readily 
metathesized. At t
8
, the rhyme pole becomes active and [
I
d
á] is output in like 
manner. 
Jakobson (1968) observed that in a remarkably large and disparate array 
of the world’s languages, from Chinese to English, the child’s first word is mama. 
Since this could not be sheer coincidence, Jakobson suggested that this uni­
versal was derived from the child’s bilabial sucking reflex. I take the infant’s 
sucking rhythm to be the prototypic syllabic rhythm generator which later 
subserves the organization of syllables into onsets and rhymes as well as met­
atheses like lought and thanguage. 
TABLE
 9.3. 
Output of figure 9.1 at t
3
. 
Phrase 
Juliet Romeo 
Romeo Juliet 
Word 
Juliet 
Romeo 
Foot (L) 
Jul 
iet 
Output 
“Romiet and Jul . . .” 

138  • 
HOW  THE  BRAIN  EVOLVED  LANGUAGE 
TABLE
 9.4. 
Output of figure 9.1 at t
4
. 
Phrase